Kernel-Based Fisher Minimum Discriminant Analysis and Face Recognition
Jingyu Yang · Jisuanji fangzhen · 2008
Linear (Fisher) discriminant analysis (LDA) is a well known and popular statistical method for feature extraction,but,due to its limitation of linearity,it fails to perform well for nonlinear problems in a lot of real-world applications,so it is necessary to extract nonlinear features. Though the conventional kernel Fisher discriminant analysis has overcome the nonlinear problems,the limitation of final eigenvectors’dimensions determined by class number still exists. To extract more effective classification information,a method of kernel-based Fisher minimum discriminant analysis was proposed. The proposed one overcomes the limitation of final eigenvectors’dimensions determined by class number. The results of experiments conducted on Yale and NUST603 face databases show the effectiveness of the proposed algorithm.